chenyang9799

train mistake

1

Happy Chinese New Year! There was a mistake in training: ``` File "UCF101_CRNN.py", line 173, in all_y_list = labels2cat(le, actions) # all video labels File "/data/chenyang/video_classification/video-classification/CRNN/functions.py", line 14, in labels2cat...

修改损失函数

7

如果我要修改损失函数CIOU,我只需要修改src/box.c的float box_ciou(box a, box b)函数吗

cls_normalizer and iou_normalizer

1

cls_normalizer and iou_normalizer对模型有什么影响？为什么要设置为1和0.5？论文里面没有提及啊。谢谢

数据集

1

您说Training set contains 16551 images, and validation set contains 4952 images. voc2007数据集好像没有这么多图片吧？还是说您用的voc2012？ 还有一点是，我下载的官方voc2007数据集，它的/ImageSets/Main/目录下已经有了train.txt和val.txt，您是用的它的现成的还是用代码生成的？